How much technical talent is there? A systematic estimate of the ML research pool among 3 million consultants

There is a significant but limited pool of ML talent outside academia: The study identified approximately 1,121 highly technical ML researchers across 403 consulting firms globally—a meaningful but constrained talent pool that is only twice the size of a single specialized training program (MATS). This suggests the technical workforce capable of advanced ML work is smaller than commonly assumed.

Current AI systems cannot yet perform advanced technical work independently: Despite recent advances in large language models, no AI model was able to pass the study's 3-day research and engineering work trial as of late 2025. This finding is significant for responsible AI discussions, as it indicates humans remain essential for complex technical decision-making in AI development.

Most consulting firms lack deep technical ML expertise: Of 2,121 organizations screened, only 403 offered broad ML consulting, and the study's rigorous methodology (combining LinkedIn analysis, LLM classifiers, and work trials) revealed that truly "highly technical" talent is concentrated in a small subset. This concentration matters for responsible AI because it means expertise for addressing safety and alignment challenges is scarce.

Work trial results suggest potential for AI safety talent identification: When 5 companies participated in the full work trial, all received at least conditional recommendations for technical AI safety work, indicating that consulting firms may harbor underutilized talent capable of contributing to responsible AI development and safety research.